Eighty days ago, Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) dropped 269 pages of artificial intelligence legislation into a capital that has never passed an AI law. The Great American Artificial Intelligence Act of 2026 still has no bill number, no committee referral and no markup on the calendar. What it has is an email address, GAAIA@mail.house.gov, where the sponsors invited the public to mail in edits.
Released June 4 by two Energy and Commerce members with Reps. Suhas Subramanyam (D-Va.), Scott Franklin (R-Fla.), Scott Peters (D-Calif.) and Erin Houchin (R-Ind.), it is the most detailed federal AI framework anyone in Congress has put on paper. Four titles: Frontier AI Governance, Workforce, Cybersecurity, and Research, Development, and International Cooperation. Despite the heft, its binding edge is deliberately narrow.
What the draft actually does
Obligations attach to "large frontier developers": companies with more than $500 million in annual revenue that have trained a model above 10^26 operations of training compute. A short list: OpenAI, Anthropic, Google, Meta, xAI. Startups, open-weight developers and academic labs are excluded by design.
Those companies would publish a "frontier AI framework" describing how they identify and mitigate severe risks, plus a transparency report at every model release or substantial modification. The draft defines catastrophic risk with unusual precision: a foreseeable risk of death or serious injury to more than 50 people, or more than $1 billion in property damage, from a model assisting with a weapon of mass destruction, running a cyberattack, or acting autonomously without meaningful human oversight.
Enforcement runs through licensed independent verification organizations, which would audit each large developer every six months with access to unredacted records, personnel and systems. Major safety incidents go to the Center for AI Standards and Innovation within 15 days, imminent risks within 24 hours. Noncompliance carries civil penalties of up to $1 million per violation per day. Whistleblowers get reinstatement and double back pay. CAISI would be codified and authorized at $100 million a year, up from roughly $15 million today. Nearly all of it sunsets in three years.
The preemption trade
So does the provision everyone is actually fighting about. The draft would bar states from enforcing laws "specifically regulating the development" of AI models for three years, while preserving laws of general applicability, common law remedies, and state authority over anything at or after deployment. Trahan says that was the price of a Republican co-lead.
"I will preempt, but only if we are setting the strongest possible federal standard," she told Tech Policy Press. "A preemption provision in a bill with no meaningful federal standard is a moratorium, and I will not sign onto that."
Her allies are unconvinced. AFL-CIO President Liz Shuler said that "while we would applaud any serious effort to regulate potentially dangerous AI technologies, any attempt to tie the hands of states in their efforts to keep working people safe is not acceptable." Public Citizen technology policy counsel J.B. Branch was sharper: "This bill strips states of their authority to respond to real harms consumers are experiencing." The ACLU and Americans for Responsible Innovation also opposed the preemption language. On release day, the co-chairs of the House Democratic Commission on AI and the Innovation Economy, Reps. Ted Lieu, Valerie Foushee and Josh Gottheimer, said the draft "does not meet the enormity of the moment" and "cannot serve as the basis for productive dialogue."
Industry likes this bill more than it liked blunter preemption efforts. Craig Albright, senior vice president for U.S. government relations at the Business Software Alliance, called it "a credible and substantive effort by Representatives Jay Obernolte and Lori Trahan to advance frontier AI model safety and strengthen public trust in AI." The Information Technology Industry Council also backed it. Cato Institute analyst Juan Londono warned that a regime built to unwind in three years "would subject this complex governance system to the whims of volatile political winds."
Why It Matters
Preemption is not abstract: the states are the only AI regulator currently issuing invoices. California AI Transparency Act duties became operative August 2. Illinois, Texas and Utah rules are in force. Colorado's rewritten AI Act and New York's insurer model-risk guidance both bite January 1, 2027. That patchwork is the real compliance regime.
The administration has tried to dismantle it by executive action instead. A December 11, 2025 order created a Justice Department AI Litigation Task Force to challenge state AI laws and directed Commerce to withhold portions of the $42.5 billion BEAD broadband program from states with laws it deems onerous. Colorado Attorney General Phil Weiser threatened to sue. Twenty-two state attorneys general and D.C., two of them Republicans, pushed back on FCC preemption efforts. None of it has suspended a single state law.
That is the case for the Obernolte-Trahan approach: only Congress can settle this, and trading a narrow development-layer freeze for enforceable federal audit and disclosure duties beats court-driven preemption with no federal floor underneath it. The case against is arithmetic. Congress has not passed privacy or kids online safety legislation in a decade, and a three-year freeze could outlast the standard that justified it.
What to Watch
The vote math is unforgiving. The Senate stripped a 10-year state AI moratorium from last year's reconciliation package 99-1. House Majority Leader Steve Scalise says he will look to Energy and Commerce Chairman Brett Guthrie, not Obernolte, for an AI vehicle. Lieu says his commission will hand a rival Democratic framework to Leader Hakeem Jeffries by year end, timed for a majority Democrats hope to win in November.
Three markers: whether the draft is formally introduced before the fall appropriations crush; whether the preemption title survives Energy and Commerce; and whether the sponsors shorten the freeze to buy Democratic votes. Sixty-five percent of Americans told the Annenberg Public Policy Center that government has done too little on AI, including 53 percent of Republicans. In an election year, that is both a reason to legislate and a reason to wait.
“I will preempt, but only if we're setting the strongest possible federal standard. A preemption provision in a bill with no meaningful federal standard is a moratorium, and I won't sign onto that.”— Lori Trahan, U.S. Representative (D-Mass.)